6.3 Judgment, Heuristics, Cognitive Biases, and Decision-Making Models

Judgment, Heuristics, Cognitive Biases, and Decision-Making Models

How do human beings evaluate probabilities, assess risks, and make choices in the face of uncertainty? Classical economic theory rested on the normative paradigm of Homo economicus—a perfectly rational decision-maker with unlimited computational capacity who systematically maximizes expected utility. Over five decades of empirical research spearheaded by Daniel Kahneman and Amos Tversky revolutionized cognitive psychology and behavioral economics by revealing that human judgment relies on heuristic shortcuts that, while efficient, produce systematic, predictable departures from normative rationality.

1. Dual-Process Theories of Cognition

Modern cognitive science conceptualizes human reasoning through dual-process theories, synthesized by Daniel Kahneman (Thinking, Fast and Slow, 2011), Jonathan Evans, and Keith Stanovich:

                                [ Sensory Input / Task Demand ]
                                               │
                 ┌─────────────────────────────┴─────────────────────────────┐
                 ▼                                                           ▼
     [ SYSTEM 1: Fast / Intuitive ]                              [ SYSTEM 2: Slow / Deliberative ]
     - Automatic, effortless, implicit                           - Controlled, effortful, explicit
     - High capacity, rapid execution                            - Capacity-limited (working memory)
     - Associative, emotional, heuristic                         - Rule-governed, logical, algorithmic
     - Default processing mode                                   - Monitors & overrides System 1 errors
     - Evolutionary older circuits (subcortical/limbic)          - Evolutionary newer circuits (dlPFC, ACC)

2. Canonical Judgment Heuristics and Cognitive Biases

In their seminal "Heuristics and Biases" research program, Amos Tversky and Daniel Kahneman demonstrated that individuals evaluating probabilities under uncertainty substitute difficult computational questions with simpler heuristic assessments:

1. The Availability Heuristic

The availability heuristic involves assessing the frequency, probability, or likelihood of an event based on the ease with which concrete instances or associations come to mind (retrieval fluency):

• Underlying Mechanism: Because frequent events are generally recalled faster than rare events, the brain uses cognitive retrieval ease as an informational proxy for statistical frequency.
• Systematic Distortions and Biases:
  • Salience and Media Vividness: Events that receive sensationalized, vivid media coverage (plane crashes, terrorist attacks, homicides, shark attacks) are retrieved with effortless fluency, leading people to drastically overestimate their statistical frequency relative to mundane, lethal hazards (diabetes, heart disease, asthma, stroke).
  • Letter Frequency Experiment (Tversky & Kahneman, 1973): When asked whether the letter K appears more frequently as the first letter of an English word or as the third letter, about two-thirds of participants (69%) judged that K is more common in the first position. In reality, English contains roughly twice as many words with K in the third position (make, bake, ask); however, generating words by their initial letter is far easier in lexical memory than searching by internal position.
  • Egocentric Allocation of Responsibility: In married couples or team collaborators, individuals estimate their personal contribution to shared tasks (cleaning, writing) at percentages that sum well in excess of 100%, because one's own efforts are vividly available in episodic memory while a partner's efforts are largely unobserved.

2. The Representativeness Heuristic

The representativeness heuristic involves estimating the probability that an entity ($A$) belongs to a specific category ($B$), or that an event was generated by a specific process, based on the degree to which $A$ resembles or matches the mental prototype of $B$:

• The Conjunction Fallacy (The "Linda Problem"; Tversky & Kahneman, 1983):
  • Paradigm: Participants read a personality sketch: "Linda is 31 years old, single, outspoken, and very bright. She majored in philosophy. As a student, she was deeply concerned with issues of discrimination and social justice, and also participated in anti-nuclear demonstrations."
  • Question: Participants rank the probability of various statements, including:
    • Statement 1: Linda is a bank teller. ($T$)
    • Statement 2: Linda is a bank teller and is active in the feminist movement. ($T \cap F$)
  • Result: Over 85% of participants (including statistically sophisticated graduate students) judged Statement 2 as more probable than Statement 1.
  • Theoretical Violation: This violates the fundamental conjunction rule of probability: the probability of the conjunction of two events can never exceed the probability of either constituent event alone ($P(T \cap F) \le P(T)$). Because the description matches the prototype of a feminist, System 1 substitutes similarity for probability.
• Base-Rate Neglect: The failure to integrate baseline prior probabilities (base rates) into probability judgments when descriptive, individuation information is provided. In the classic Lawyers and Engineers experiment (Kahneman & Tversky, 1973), participants were informed of a pool containing 70 lawyers and 30 engineers (or vice versa). Participants given a stereotypical engineer description produced nearly the same probability estimates whether the pool contained 70 engineers or 30 engineers, largely ignoring the base rates.
• Gambler's Fallacy and Misconception of Chance: The erroneous conviction that independent, identically distributed random events are self-correcting. For example, after observing a fair coin land on "Heads" five consecutive times, individuals predict that "Tails" is "due" on the sixth flip. People expect short sequences of random events to locally reflect the global properties of the generating mechanism (Law of Small Numbers).
• Neglect of Regression Toward the Mean: When an extreme performance or measurement is recorded, subsequent measurements will naturally tend to regress toward the statistical population average purely due to measurement error and stochastic variance. Failure to appreciate this leads to erroneous causal attributions (e.g., flight instructors concluding that severe criticism improves subsequent student landing performance, when the improvement was merely statistical regression from an unusually poor landing).

3. Anchoring and Adjustment Heuristic

The anchoring and adjustment heuristic occurs when individuals estimate an unknown numerical quantity by starting from an initial baseline value (the anchor) and making adjustments to arrive at a final answer:

• Mechanism: Adjustments away from the anchor are systematically insufficient, leaving the final estimate heavily biased toward the anchor value.
• Arbitrary and Irrelevant Anchors: Anchoring occurs even when the starting value is overtly arbitrary, transparently random, or irrelevant. In Tversky and Kahneman's (1974) classic demonstration, participants observed a wheel of fortune rigged to stop at either 10 or 65, and were then asked to estimate the percentage of African nations in the United Nations. Participants who saw the anchor 10 gave a median estimate of 25%, whereas those who saw 65 gave a median estimate of 45%.
• Real-World Impact: Anchoring powerfully distorts legal sentencing, real estate negotiations, commercial pricing strategies, and clinical diagnostic impressions.

3. Normative vs. Descriptive Decision Models: Expected Utility vs. Prospect Theory

To understand human decision making, cognitive psychology draws a rigorous contrast between normative models of idealized rationality and descriptive psychological models of actual behavior:

Normative Models: Expected Value and Expected Utility Theory

• Expected Value (EV): The mathematical average outcome of an uncertain gamble: $EV = \sum p_i x_i$. EV fails to explain why people purchase insurance or decline fair gambles.
• Expected Utility Theory (EUT; John von Neumann & Oskar Morgenstern, 1944): Replaces objective monetary values ($x_i$) with subjective psychological utility ($U(x_i)$): $EU = \sum p_i U(x_i)$.
  • Assumes rational actors obey key axioms: Transitivity (if $A \succ B$ and $B \succ C$, then $A \succ C$), Completeness, and Independence.
  • Assumes diminishing marginal utility of wealth, yielding a globally concave utility curve that explains general risk aversion.

Descriptive Behavioral Paradigm: Prospect Theory (Kahneman & Tversky, 1979)

Kahneman and Tversky introduced Prospect Theory (for which Kahneman won the 2002 Nobel Memorial Prize in Economic Sciences) to describe how humans actually make choices under risk and uncertainty. Prospect Theory departs from Expected Utility Theory through two revolutionary mathematical/psychological components:

                   Value V(x)
                        │          /  (Concave for Gains -> Risk-Averse)
                        │         / 
                        │       /   
       Losses           │     /       Gains
  ──────────────────────┼─────────────────────── Outcome x
             \          │  (Reference Point = 0)
              \         │
               \        │
                \       │  (Convex for Losses -> Risk-Seeking)
                 \      │
                  \     │  (Steeper Slope: Loss Aversion ~2:1)

1. The Value Function $v(x)$

• Defined on Reference Points: Subjective value is computed not in terms of absolute total wealth states, but as deviations (gains and losses) relative to a neutral subjective reference point (typically the current status quo).
• Asymmetric Curvature:
  • Domain of Gains (Concave): Exhibits diminishing sensitivity for gains ($v''(x) < 0$ for $x > 0$). Moving from $0 to $100 yields greater psychological utility than moving from $1,000 to $1,100. Induces Risk Aversion for Gains (e.g., people strongly prefer a sure $500 over a 50% gamble for $1,000).
  • Domain of Losses (Convex): Exhibits diminishing sensitivity for losses ($v''(x) > 0$ for $x < 0$). Moving from losing $0 to -$100 hurts far more than moving from -$1,000 to -$1,100. Induces Risk Seeking for Losses (e.g., people prefer a 50% chance of losing $1,000 over a guaranteed, sure loss of $500).
• Loss Aversion: The value function is significantly steeper in the domain of losses than in the domain of gains ($|v(-x)| > v(x)$). Formally, the loss aversion coefficient $\lambda \approx 2.0 \text{ to } 2.5$. "Losses loom larger than gains": the psychological pain of losing $100 is roughly double the pleasure of gaining $100.
  • Endowment Effect (Richard Thaler, 1980): People assign substantially higher value to goods they own than to identical goods they do not own. In classic experiments, participants randomly gifted a coffee mug demanded roughly twice as much money to sell it ($WTA$) as non-owners were willing to pay to acquire it ($WTP$).

2. The Probability Weighting Function $\pi(p)$

• Decision makers do not multiply value by objective probabilities ($p$), but by non-linear subjective decision weights ($\pi(p)$):
  • Overweighting of Small Probabilities: Rare, low-probability events ($p < 0.05$) are psychologically overweighted ($\pi(p) > p$). This accounts for the simultaneous purchase of lottery tickets (risk-seeking for low-probability massive gains) and disaster insurance (risk-averse for low-probability catastrophic losses).
  • Underweighting of Moderate and High Probabilities: Moderate and high probabilities are psychologically underweighted ($\pi(p) < p$).
  • The Certainty Effect: A reduction in probability from 100% to 99% induces a massive psychological penalty out of proportion to a reduction from 40% to 39%.

4. Framing Effects and Pervasive Cognitive Fallacies

Framing Effects: The Asian Disease Problem

Tversky and Kahneman (1981) demonstrated that subtle changes in the semantic presentation or "framing" of options alter the subjective reference point, causing choices to flip predictably between risk aversion and risk seeking:

• The Scenario: The US is preparing for an outbreak of an unusual Asian disease expected to kill 600 people. Two alternative intervention programs are proposed:
  • Gain Frame (Lives Saved):
    • Program A: 200 people will be saved. [72% chose A: Risk-Averse]
    • Program B: 1/3 probability that 600 people will be saved, and 2/3 probability that no one will be saved. [28% chose B]
  • Loss Frame (Deaths / Lives Lost):
    • Program C: 400 people will die. [22% chose C]
    • Program D: 1/3 probability that nobody will die, and 2/3 probability that 600 people will die. [78% chose D: Risk-Seeking]
• Theoretical Implication: Mathematically, Program A is identical to Program C (both result in 200 survivors and 400 deaths), and Program B is identical to Program D. However, the gain framing casts choices in terms of lives saved, triggering the concave value function and risk aversion, whereas the loss framing casts choices in terms of deaths, triggering the convex value function and risk seeking.

Pervasive Decision Fallacies

• Sunk Cost Fallacy: The irrational tendency to continue an endeavor, project, or investment once resources (money, time, effort) have been committed, even when the future marginal costs exceed marginal benefits (e.g., forcing oneself to finish a terrible movie in a theater simply because the ticket was expensive; the Concorde supersonic airliner project). Rational choice dictates evaluating decisions purely based on future expected utility.
• Hindsight Bias ("I knew it all along" effect; Baruch Fischhoff, 1975): The retrospective tendency to exaggerate one's past ability to have anticipated an outcome once that outcome has already occurred. Driven by reconstructive memory: outcome knowledge is absorbed into the recalled prior judgment.
• Overconfidence Effect and Calibration: Subjective confidence systematically exceeds objective accuracy across a wide range of declarative knowledge tasks. Experts often exhibit the most pronounced overconfidence when forecasting complex future geopolitical or economic developments.
• Dunning-Kruger Effect (Justin Kruger & David Dunning, 1999): A metacognitive deficit wherein individuals with low competence or expertise in a specific domain suffer a dual burden: they arrive at erroneous conclusions and make unfortunate choices, but lack the metacognitive sophistication required to recognize their own incompetence. Conversely, high-performing experts tend to underestimate their relative ability, erroneously presuming tasks that are easy for them are equally easy for others.

5. Bounded Rationality and Satisficing: Herbert Simon

Long before the heuristics-and-biases program emerged, Nobel laureate Herbert Simon (1955, 1957) challenged the classical economic construct of Olympian rationality:

      [ Classical Maximizing / Optimization ]            [ Bounded Rationality / Satisficing ]
      ───────────────────────────────────────            ─────────────────────────────────────
      - Assumes infinite computational capacity          - Recognizes human cognitive & memory limits
      - Exhaustive search of all alternatives            - Heuristic search terminated at threshold
      - Maximizes global mathematical utility            - Chooses first option meeting aspiration level
      - High regret, paralysis, dissatisfaction          - Adaptive, computationally efficient, content

• Bounded Rationality: Human decision-makers are bounded by severe internal cognitive constraints (finite working memory capacity, limited attentional resources, perceptual bandwidth) and external environmental constraints (imperfect information, severe time pressure).
• Satisficing (Portmanteau of Satisfy and Suffice): Instead of conducting an exhaustive algorithmic search across all possible alternatives to identify the optimal choice (maximizing), human actors set an internal aspiration level of acceptability. They evaluate available options sequentially until discovering the first option that meets or exceeds this aspiration threshold, immediately terminating the search.
• The Paradox of Choice (Barry Schwartz, 2004): Modern empirical studies demonstrate that while maximizers achieve objectively marginally superior outcomes (e.g., obtaining a slightly higher starting salary), satisficers consistently report significantly higher subjective well-being, lower post-decisional regret, less counterfactual rumination, and greater life satisfaction.